Datasets:
Modalities:
Geospatial
Languages:
English
Size:
1K<n<10K
Tags:
remote-sensing
building-height-estimation
geospatial
high-resolution-imagery
benchmark
building-segmentation
License:
metadata
license: other
language:
- en
task_categories:
- image-segmentation
- object-detection
tags:
- remote-sensing
- building-height-estimation
- geospatial
- high-resolution-imagery
- benchmark
- building-segmentation
- aerial-imagery
- overhead-imagery
- coco
- tiff
pretty_name: HighBuild-1M Small Sample
size_categories:
- 1K<n<10K
HighBuild-1M Small Sample
This repository provides a small inspection subset of HighBuild-1M for reviewers.
The full benchmark is hosted at:
https://huggingface.co/datasets/feifei140729/HighBuild-1M
Purpose
This small sample is provided for quick inspection of data quality, file organization, spatial alignment, and annotation format. It is not intended to replace the full benchmark or reproduce all reported benchmark results.
Contents
The sample follows the same triplet structure as the full HighBuild-1M dataset:
images/: RGB 1024 x 1024 overhead image tiles.masks/: spatially aligned float32 TIFF building-height masks. Non-building pixels are 0, and building pixels store building-height values in metres.coco_json/: COCO-style building instance annotations, including polygons, bounding boxes, building-height attributes, and estimated-height flags.
Files with the same stem across the three directories correspond to the same sample.
File Organization
small-sample/
├── images/
├── masks/
├── coco_json/
├── README.md
└── .gitattributes